193
British Journal of Social Psychology (2011), 50, 193–215
C 2010 The British Psychological Society
The
British
Psychological
Society
www.wileyonlinelibrary.com
The power of talk: Developing discriminatory
group norms through discussion
Laura G. E. Smith1∗ and Tom Postmes2
1
2
The University of Queensland, St Lucia, Australia
The University of Groningen, The Netherlands
Research has shown that group discussion can increase intergroup prejudice and
discrimination. However, we know little about the process by which discussion has this
effect. Therefore, four studies were conducted in a real-world context to investigate
this process. Results suggest that discussing a negative societal stereotype (relative to
individual rumination in Studies 1 and 3 and alternative discussions in Studies 2 and
3) increases intentions to engage in discrimination against the out-group target of the
stereotype. This is mediated by the formation of an in-group norm which supports
discrimination (Study 1) and the extent to which the discussion validates the stereotype
(Study 2). A fourth study manipulated the extent to which consensus on the negative
stereotype was reached through discussion. When the discussion ended in consensus,
participants have greater intention to undertake collective action against the stereotyped
out-group, mediated by a congruent in-group norm. These results provide evidence that
the process by which discussion increases intergroup discrimination is via the formation
of discriminatory local group norms.
Group discussion can have profound behavioural consequences (e.g., Lewin, 1953). It
can lead individuals to accept higher levels of risk (Fraser, Gouge, & Billig, 1971; Stoner,
1968), and polarize their opinions and attitudes (e.g., Moscovici & Zavalloni, 1969; for
a review, see Turner, 1991). Alarmingly, research has shown that group discussion can
also increase intergroup prejudice (Myers & Bishop, 1970) and discrimination (Smith &
Postmes, 2009). However, we have limited understanding of why discussion had this
effect in these studies. Accordingly, it is important to examine the processes by which
discussion increases discriminatory behaviour in the real world. In this article, we draw
upon insights from the social identity tradition and small group research to demonstrate
that the process by which discussion changes behavioural intentions is via the formation
of local group norms. Moreover, we aim to demonstrate that these norms form if the
discussion supports the perception that others share similar views of the out-group,
making them subjectively valid world-views.
∗ Correspondence
should be addressed to Dr Laura G. E. Smith, School of Business, The University of Queensland, St Lucia,
Qld 4072, Australia (e-mail: [email protected]).
DOI:10.1348/014466610X504805
194
Laura G. E. Smith and Tom Postmes
Understanding intergroup behaviour and in-group norms has been central to the
social identity approach since social identity theory (SIT) was first articulated in the
1970s (Tajfel & Turner, 1979). Building on SIT, self-categorization theory (Turner, Hogg,
Oakes, & Reicher, 1987) argued that group members’ behaviour is driven by their
collective understanding of the intergroup context. It appears therefore that norms form
if group members make common deductions about appropriate social behaviour in a
particular context. Sherif’s (1956) boys’ camp studies are often cited as an example
of this type of norm formation, whereby a competitive intergroup context nurtured
norms for intergroup competition and conflict. For many years, this has been central
to our understanding of how collective behaviour originates. However, individually
deducing in-group norms from the intergroup context may not be the only method
by which these norms develop. Research demonstrates that group discussion can also
significantly change attitudes and behaviour. This suggests that norms for collective
behaviour may also develop through small group interaction (Postmes, Haslam, &
Swaab, 2005). Of course, the only way in which group members can know that their
deductions about the intergroup context are shared within the group is through some
form of communication. Discussion is thus necessary for the coordination of group
behaviour.
Evidence for the impact of discussion on group behaviour is compelling. For
example, Myers and Bishop (1970) divided participants into groups based on their
pre-measured racial attitudes. They found that participants who were already inclined
towards prejudice became more so through discussion (and vice versa). This provides
a dramatic demonstration of the power of talk to produce a group shift in attitudes.
Although Myers and Bishop (1970) did not measure the subsequent behaviour of their
participants, findings from recent research suggest that a group change in attitude after
discussion may lead to a change in associated group behaviour. For example, Smith and
Postmes (2009) found that discussion led to an increase in discrimination that could
be explained by an increase in a hostile intergroup norm. In other words, discussion
impacted on the group norm, and the group norm impacted on the group’s behaviour.
It appeared that the norm thus constituted the driving process by which discussion
achieved its impact on group behaviour.
However, there is currently a paucity of research which examines how these norms
for intergroup discrimination may develop through interaction. One interesting insight
into this process arose from Stott and Drury’s (2004) research. They found that discussing
an out-group stereotype increased participants’ support for collective protest. Therefore,
it appeared that discussing the stereotype had a direct impact on collective behaviour.
Indeed, many authors have argued that stereotypes ignite intergroup discrimination
because they reflect values, ideologies, and a particular world view (e.g., Jost & Banaji,
1994). For example, historians have named the anti-Semitic stereotype as a contributing
factor of German complicity with the Nazi regime (see Gregor, 2005). Although Stott
and Drury (2004) report no evidence that discussion had this effect because it changed
in-group norms, it seems that discussing stereotypes may be an important part of the
psychological machinery which coordinates group behaviour. As we describe below,
the discovery that group members share opinions about the out-group can result in the
perception that those opinions are a valid basis for behaviour. The communication of
stereotypic traits (which is guided by multiple processes; for reviews, see Kashima,
Fiedler, & Freytag, 2008) is an important way in which this group consensus can
emerge.
The power of talk
195
On their own, individuals may feel uncertain that a societal stereotype of the out-group
is valid. However, discussion provides an opportunity for those individuals to assess the
extent to which their individual beliefs are similar to those of other group members and
often leads to an increase in group consensus on the stereotypical traits (e.g., Haslam,
1997; Haslam, Oakes, Reynolds, & Turner, 1999; Ruscher, Hammer, & Hammer, 1996;
Thompson, Judd, & Park, 2000). If discussion results in a high degree of consensus
about the out-group, this can increase individuals’ confidence about the validity of the
stereotype (as argued by Festinger, 1950, 1954; and in a different way by McGarty,
Turner, Oakes, & Haslam, 1993; Turner, 1991). The group’s increased perceptions of
the validity of the stereotype may provide a basis for subsequent intergroup behaviour
(Haslam, Turner, Oakes, McGarty, & Reynolds, 1998; Reicher, Hopkins, & Condor, 1997;
Stangor, Sechrist, & Jost, 2001). Therefore, discussion of a negative out-group stereotype
could lead to an increase in negative intergroup behaviour targeted towards that outgroup. Importantly, this would represent a group shift, and therefore indicate a change
in the local group norm.
In sum, there is a growing body of research which suggests that sharing individual
opinions and stereotypes can influence wider group processes (e.g., Haslam, 1997;
Postmes, Haslam, et al., 2005; Postmes, Spears, Lee, & Novak, 2005), but we are as
yet unclear on how this occurs. Unravelling this process may enable us to examine
in greater detail how group action develops from the meeting of individuals. The
present research was designed to provide evidence for the process in a real-world
context by systematically examining the effects of the small group discussion of a
wider societal stereotype. This research is distinguishable from previous research on
norm formation (such as Sherif, 1956) because it includes individual rumination as an
experimental control. This allowed us to compare the consequences of discussion to
those of individual rumination and demonstrate the unique impact that discussion has
on norms and collective behavioural intentions.
The present research
In four studies, we examined consequences of in-group members (Britons) discussing
out-group (immigrant) stereotypes. At the time of the present research, immigration
was at the centre of an ongoing debate in Britain. There was a popular concern about
numbers of immigrants in Britain (Ipsos MORI, 2008). Support for the UK Independence
Party and the British National Party had increased in recent years, and these parties’
call for zero net immigration was widely publicized. The politically conservative media
tended to focus on the supposed economic and cultural threats of immigration for Britain
(e.g., Daily Mail, 2009).
STUDY 1
Study 1 was an exploratory study that examined the impact of group discussion about
immigrants on endorsement of pro- or anti-immigration policies. Using British citizens
as the target in-group and face-to-face interaction, the first study simply compared
the products of an in-group discussion about the out-group (immigrants) with a no
discussion control condition in which individuals reflected about the out-group alone.
We hypothesized (a) that if norms form through dialogue, then group discussion would
have a different effect on behavioural intentions than individual reflection, when the
196
Laura G. E. Smith and Tom Postmes
same identity and context is salient. We predicted (b) that the emergent norm would be
positively related to intergroup behaviour, and would mediate the difference between
the two experimental conditions on behavioural intentions. We assumed that if the
majority of participants in our samples were opposed to immigration in line with
the societal stereotype described above, participants would become more prejudiced
following group discussion. The first study tested this assumption.
Method
Participants and design
Participants were 39 British students (mean age = 17.0, 56% female) from a state school
in the UK, recruited during an educational visit to their establishment. The study had a
two-condition (type of reflection: group discussion vs. individual reflection) betweensubjects design. Participants were randomly allocated to N = 7 groups of three in the
discussion condition and N = 6 groups of three in the individual rumination condition.
All were unpaid volunteers and naive about the purpose of the study. Parental/guardian
(opt-out) consent was requested in advance.
Procedure
For 5 minutes, participants either engaged in a group discussion in their small group of
three people, or individually reflected, on the out-group (immigrants). The instructions
were either ‘As a group’ or ‘On your own’, ‘Formulate an impression of immigrants.
What do you think immigrants are like as a group of people? What do they do? Who
are they? Where do they come from? What is their relation to British people?’ The
wording of this instruction required participants to think in general group terms, so
that their subsequent responses were driven by their pre-existing cultural knowledge
of the out-group stereotype. Each participant was asked to individually record the top
five points that arose during discussion or during their individual rumination. They were
then asked to complete a questionnaire individually and in silence. After completing
the standardized scales, participants were debriefed in person by the experimenter and
provided with a full written debriefing.
Dependent measures1
Participants’ written description of the out-group was used as a measure of qualitative
stereotype content. The main quantitative dependent measure was a three-item support
for intergroup discrimination scale (␣ = .67), which asked, ‘How much social security
money should immigrants get?’, ‘How much free healthcare should be provided for
immigrants?’, and ‘How much should immigrants get in benefits?’. Items were answered
on a seven-point Likert type scale (1 = Much more than British people, 4 = Equal,
7 = Much less than British people). This scale was designed to measure support for
discrimination towards the out-group, but was able to capture support for both pro- and
anti-out-group policies. All other scales consisted of statements with which participants
1 Space
constraints prevent us from reporting all dependent variables here. All four studies included measures of collective
efficacy, social support, and anger. There were no significant condition effects or significant mediation for these variables,
with one exception: In Study 1, anger mediated the condition effect together with group norm. However, this effect was not
replicated in subsequent studies.
The power of talk
197
indicated agreement (1 = ‘strongly disagree’, to 7 = ‘strongly agree’). A four-item scale
measured perceptions of the in-group norm (␣ = .72): ‘We believe that keeping Britain
British is the right thing to do’; ‘For students like us it is normal to favour our own kind’;
‘Students like us believe that it’s right to treat immigrants better than British people’
(reverse coded); and ‘The average sixth form student believes that immigration enriches
British society’ (reverse coded).
To gauge the perception of consensus with the other group members, shared
cognition was measured via a four-item scale (␣ = .77). This scale included the
instruction, ‘When the statement refers to “we”, this means you and the other students’.
Items were, ‘We agree with each other about the issues’; ‘We have similar ideas’, ‘We
are on the same wavelength’, and ‘We are aware of the differences between our views’
(reverse coded).
An adaptation of Guimond and Dambrun’s (2002) generalized prejudice scale (10
items, ␣ = .83) measured attitudes towards immigrants. Items included, ‘The entry of
foreign families into Britain should be more stringent’ and ‘Immigrants should not be
given responsibility or positions of authority over the British’. A seven-item adaptation
of Ellemers, Kortekaas, and Ouwerkerk’s (1999) measure of identification was included
(␣ = .51) to measure identification with the other students who participated in the
discussion groups or individual rumination, with sample items, ‘I identify with the other
students’ and ‘I feel good about these students’. Identification was included to examine
whether group discussion increased social identification and thereby galvanized social
identity. Inter-scale correlations suggested that the discriminant validity of the scales was
acceptable (Table 1).
Table 1. Mean individual-level scores and scale intercorrelations in the group discussion (N = 21) and
individual reflection conditions (N = 18), Study 1
Analytic strategy
In Studies 1–3, analyses were conducted with hierarchical linear modelling (HLM) in
HLM 6.03 (Raudenbush & Bryk, 2002). This is appropriate when individuals are nested
within groups. A priori predictions were tested by means of dummy and contrast
variables, following the procedure outlined by Cohen, Cohen, West, and Aiken (2003).
198
Laura G. E. Smith and Tom Postmes
In Study 1, we tested two hypotheses: (a) whether group discussion increased support
for intergroup discrimination relative to individual reflection and (b) whether the group
norm mediated this condition effect.
Content analyses were performed on each individual’s written descriptions of the
out-group and identified (1) whether the participants stereotyped the out-group and/or
explicitly rejected the notion of stereotyping them; (2) the content of the stereotype;
(3) the affective tone of the stereotype; and (4) perceptions of the intergroup context.
Iterative inductive and deductive coding procedures were followed by the primary coder,
who was not blind to the hypotheses (Miles & Huberman, 1994). A blind second coder
independently coded the data. Inter-rater agreement was 95%, and inter-rater reliability
was excellent: Cohen’s (1960) kappa ranged from ␬ = .71 to 1.00 across codes. Codes
are described in Table 2. The frequency with which each code occurred in the data was
recorded and the frequency count in each condition was compared across individuals
nested within groups using a Bernoulli sampling HLM analysis with a logit link function.
Results
Content analysis
Participants in both conditions almost universally stereotyped the out-group (Table 2).
Notably, there was a significant difference between the conditions in the frequency
with which the welfare fraud code was mentioned, ␥ = −2.35, log-odds = .10, p =
.05; and marginal differences were found on the codes for racial categorization, ␥ =
−2.43, log-odds = .09, p = .06; and intergroup tensions, ␥ = −1.94, log-odds = .14, p =
.06. A relatively greater number of participants mentioned these codes after discussion
than individual rumination (Table 2). Beliefs about welfare fraud and intergroup tension
appeared to be related. The opinions that, ‘They think they could live over this country
and take our money’ and similarly, ‘Some immigrants are merely here to deal drugs
and rake money off the state’, illustrated the intergroup tension that appeared to arise
for participants who articulated the argument that immigrants engaged in welfare fraud.
Welfare fraud appeared to be a focal aspect of the out-group stereotype in the discussions,
with 88% of the participants mentioning the issue, compared to 30% of participants in
the individual rumination condition. No other codes differed significantly in frequency
between the two conditions (all p’s > .08).
Analysis of standardized scales
Analyses showed that there was significantly more support for intergroup discrimination
in the discussion condition than the individual rumination condition, ␥ = .50, p = .04
(Table 1). There was also a significant group norm main effect, ␥ = 1.35, p = .03, with
more normalization of intergroup discrimination in the discussion condition than the
individual condition. There was a main effect for shared cognition, with a perception
of more shared cognition after discussion than individual reflection ␥ = 1.10, p =
.024. Condition had no effect on generalized prejudice (p > .20), although prejudice,
discrimination, the norm, and shared cognition were significantly positively correlated
(Table 1). There was no difference in overall social identification or on the subscales
between the two conditions.
Table 2. Content analysis in the discussion (N = 21) and individual rumination (N = 18) conditions, Study 1
The power of talk
199
Laura G. E. Smith and Tom Postmes
Table 2. (Continued)
200
The power of talk
201
Mediation2
In order to test whether the norm mediated the effect of the conditions on support for
intergroup discrimination, we entered the norm into the multi-level model, predicting
endorsement of intergroup discrimination alongside the condition variable. The slope
for the group norm was significant ␥ = .20, p < .001. On entering this mediator into
the model, the difference between the discussion condition and the individual condition
became non-significant, ␥ = .23, p = .33 suggesting that the norm mediated the condition
effect, Sobel z = 2.01, p = .04 (Figure 1).
Discussion
This exploratory study showed that group discussion (relative to individual rumination)
focused more on specific prejudiced elements of the out-group stereotype and intergroup tension, and led to more support for discrimination against the out-group. This
result was mediated by perceptions of an in-group norm which supported intergroup
discrimination. There was also a greater perception of shared cognition within the group
after discussion compared to individual rumination, indicating more consensus on the
issue of immigration. Therefore, it appeared that discussion uniquely contributed to the
formation of local norms, and it was through these norms that discussion impacted on
support for intergroup discrimination.
An alternative explanation for the findings of Study 1 was that discussion increased
identity salience, and this salience led to an increase in intergroup hostility. This seems
unlikely, as participants in both conditions were instructed to think or discuss the
same topic. Therefore, the salience of British and immigrant identities should have
been constant across the conditions. This alternative explanation was explored in
Study 2.
STUDY 2
The effects we found in Study 1 may be accounted for by two distinct processes. First,
norm formation may have occurred because the out-group stereotype was validated
through discussion. Alternatively, norm formation may have occurred because the
discussion increased identity salience (cf. Postmes, Spears, et al., 2005). To test whether
group discussion per se may increase identity salience which leads to the inference of
shared norms, groups also interacted in the control condition in Study 2, but on an
unrelated topic.
We had limited control over the content of discussions in Study 1. It is possible that
participants may have discussed potential actions or policies towards the out-group in
addition to the out-group stereotype. Therefore, in Study 2 we added a condition in
which participants were explicitly asked to discuss intergroup actions. This meant we
could assess the degree to which discussing the out-group stereotype in particular or the
out-group plus intergroup actions was the driving process behind the results of Study 1.
As identity salience should be equal in the two conditions in which participants discussed
2 We
used the multi-level first-order Taylor series approximation to estimate the standard error of the mediated effect as
is recommended for multi-level mediation with a Level-2 predictor, Level-1 mediator, and Level-1 outcome (see Krull &
MacKinnon, 1999).
202
Laura G. E. Smith and Tom Postmes
Norm
perception
1.35*
Group discussion vs.
Individual reflection
.20**
(.50*)
.23, ns
Intergroup
discrimination
Figure 1. Coefficients for the mediation model, Study 1 (∗ p < .05; ∗∗ p < .01).
the out-group, these conditions provided a further test of the alternative explanation for
the Study 1 results.
Based on the above analysis, three conditional hypotheses were tested: (1) if reaching
consensus on discriminatory intergroup policies was the driving process behind an
increase in support for discriminatory policies, then the discussion of actions would be
likely to increase action intentions relative to the conditions in which participants did
not discuss action. However, (2) if discussing the stereotype was the driving process,
then discussing the stereotype may be equally or more likely to increase action intentions
than discussing actions themselves. Furthermore, (3) if salience was the driving process,
then there should be no difference between the conditions. Finally, (4) if norm formation
occurred via a process of social validation, the effects of condition on action intentions
would be mediated by subjective social validation.
Method
Participants and design
Participants were 78 White British sixth form students who volunteered from a different
state comprehensive school to Study 1, and were recruited during an educational visit.
The mean age was 17.0 years, 31 participants were female. The study had a threecondition (topic of interaction: out-group stereotype vs. out-group stereotype plus a
plan for social action vs. irrelevant control) between-subjects design. Participants were
randomly allocated to groups of three people within each condition (N = 10 groups in
the stereotype condition, N = 11 in the stereotype plus action condition, and N = 5 in
the irrelevant control condition3 ). All were unpaid and naive as to the purpose of the
study. Consent and debriefing arrangements were identical to Study 1.
Procedure
Participants engaged in a 5-minute discussion. In the out-group stereotype condition,
instructions were identical to the discussion condition of Study 1. In the stereotype
plus action condition (hereafter referred to as the action condition), instructions
included, ‘[Formulate] a plan to combat any problems immigration causes’. In the
control condition, participants discussed an irrelevant topic (whether or not the British
monarchy is outdated).
3 Groups
were distributed unequally across conditions due to logistical constraints in recruiting participants.
The power of talk
203
Dependent measures
Intergroup action intentions were measured with a four-item scale that asked whether or
not participants were prepared to engage in concrete and direct action about immigration
(␣ = .76). Participants responded either 1 = ‘No’, or 2 = ‘Yes’ to the questions, ‘I agree
to participate in a demonstration about immigration policy in Britain’, ‘I agree to have my
name added to a petition to make the government listen to my group’s views’, ‘I would
do something with fellow group members to address the problems that immigration
brings to Britain’, and ‘I would participate in raising our collective voice to address the
problems that immigration brings to Britain’.
A three-item scale measured subjective social validation (␣ = .79): ‘The discussion
made me feel that my views are more legitimate’; ‘My beliefs about immigration are
justified’; and ‘I feel that my views on immigration are well-founded’. The purpose of
including this scale was to compare the extent to which individuals perceived social
validation of their views about the out-group across conditions. Participants indicated
agreement with these and all subsequent items on seven-point Likert-type scales (1 =
‘strongly disagree’, 7 = ‘strongly agree’). A single item measured participants’ certainty
about future courses of action: ‘I have a clear idea about where things need to go with
immigration in this country’. The generalized prejudice scale was the same as in Study 1
(10 items, ␣ = .88). A single item measured social identification: ‘I identify with the other
members of my group’. Inter-scale correlations (Table 3) suggested that the discriminant
validity of scales was acceptable.
Table 3. Mean individual-level scores and inter-scale correlations in the stereotype (N = 30), action
(N = 33), and control (N = 15) conditions, Study 2
Results
The effects of experimental condition on action intentions were explored in the initial
model, ␹ 2 (24) = 49.71, p < .001. Action intentions were significantly lower in the
irrelevant discussion condition compared to the out-group stereotype condition, ␥ =
0.93, p < .001 (Table 3) but significantly higher after discussion of the stereotype, than
after discussion of the stereotype plus courses of action, ␥ = 1.13, p = .01. Intentions in
204
Laura G. E. Smith and Tom Postmes
Social
validation
.51*
Stereotype discussion
vs.
Irrelevant discussion
.36**
(.93**)
.77**
Intergroup
action
intentions
Figure 2. Coefficients for the mediation model, Study 2 (∗ p < .05; ∗∗ p < .01).
the irrelevant discussion condition were not significantly different to those in the action
condition ␥ = 0.52, p = .37.
The effects of condition on social validation were then explored, ␹ 2 (24) = 37.93,
p = .03. Participants felt more validated in the stereotype condition than the irrelevant
condition, ␥ = 0.51, p < .01, and in the action condition than the irrelevant condition
␥ = 0.57, p = .04. Although the mean was higher in the stereotype condition than the
action condition, the difference was not quite significant, ␥ = 0.47, p = .11. As in Study
1, there were no condition effects for identification or generalized prejudice (all p’s >
.08). There was somewhat more certainty about action in the stereotype condition than
the irrelevant condition, although not significantly so, ␥ = 0.38, p = .10. Notably, there
was significantly less certainty in the action condition than in the irrelevant condition,
␥ = 0.57, p = .04. There was also less certainty in the action condition than when groups
discussed the stereotype, ␥ = 0.95, p = .03. Examination of intra-class correlations
between groups in each condition indicates that there was less variability in action
intentions between groups in the stereotype condition ICC1 = 15.87 than the action
condition ICC1 = 26.86 (and the highest variability was found in the irrelevant condition
ICC1 = 37.14).
Mediation
Validation showed the same pattern of between-condition differences as action intentions, and the relationship between validation and action intentions was marginally
significant, ␥ = 0.29, p = .06. Thus, validation was included in the model as a mediator.
On entering this mediator into the model, the slope for validation was significant,
␥ = 0.36, p < .001. The difference between the stereotype and the irrelevant discussion
conditions reduced from ␥ = 0.93, p < .001 to ␥ = 0.77, p = .01. Thus, validation
partially mediated the effect of experimental condition on action intentions; Sobel z =
2.24, p = .03 (Figure 2).
Discussion
In Study 2, the topic of interaction was varied to distinguish the effects of interaction
per se from the effects of discussing out-group stereotypes only, or discussing outgroup stereotypes as well as potential action towards that group. The lowest action
intentions scores occurred in the irrelevant (control) condition compared with the
experimental conditions. Participants in the control condition also scored lower on
The power of talk
205
validation. Therefore, we concluded that interaction with group members per se did not
lead to greater action intentions.
It was also hypothesized that if reaching consensus on discriminatory intergroup
policies was the driving process behind an increase in support for discriminatory policies,
then the discussion of actions would be likely to increase action intentions relative to
the conditions in which participants did not discuss action. However, discussing actions
resulted in lower action intentions, and lower levels of social validation than discussing
only the out-group stereotype. Instead, the results supported the argument that social
validation of the out-group stereotype was the driving process behind the effects. It
seems unlikely that an increase in identity salience could have significantly affected the
results, as this was constant across the conditions in which participants discussed the
out-group.
The data suggested that there was greater action certainty and less within-condition
variability on action intention when the groups discussed an out-group stereotype than
a concrete plan of action. Therefore, it appeared that discussions about courses of
action reduced certainty about action compared with discussion of only the stereotype.
Retrospectively, as with any real-world situation, there may have been difficulties in
reaching an explicit, consensual plan for action in this context in this condition.
Participants may have been unwilling to explicitly endorse the political position of
taking action against a lower status out-group.
Social validation appeared to be an important psychological product of intragroup
interaction and mediated the effects of condition on action intentions. Thus discussions
that validated participants’ individual views on the out-group stereotype appeared to
provide a firmer foundation for their intentions to engage in intergroup action. This result
provides support for the argument that group discussion can be a source of validation
and action confidence (Baron et al., 1996; Luus & Wells, 1994; Petty, Briñol, & Tormala,
2002). Overall, these results provide further support for the idea that discussions
about an out-group stereotype can aid the development of intentions for collective
behaviour.
STUDY 3
This study combined the designs of the previous two studies, by independently
manipulating the type of reflection on the out-group (group discussion vs. individual
reflection) and the topic of reflection (out-group stereotype only vs. out-group stereotype
plus action). The aim was to replicate the previous findings with a behavioural measure
of hostile social action. We hypothesized that (a) overall, there would be more hostile
action after group discussion than after individual reflection and (b) there would be
a simple main effect between the two discussion conditions: participants would show
more hostility towards the out-group when they discussed the out-group stereotype than
when they discussed the stereotype plus specific actions.
Method
Participants and design
Participants were 78 British students (mean age = 16.7 years; 77% female) recruited
during an educational visit to a different state school to Studies 1 and 2. The study had a
206
Laura G. E. Smith and Tom Postmes
2 (type of reflection: group discussion vs. individual reflection) × 2 (topic of reflection:
out-group stereotype vs. out-group action) between-subjects design. Participants were
randomly allocated to groups of three. There were N = 7 groups in each of the stereotype
conditions, and N = 6 groups in each of the action conditions. Consent and debriefing
arrangements were the same as for the previous studies.
Procedure
Participants were asked to either undertake a 5-minute discussion in small groups or
reflect on the specified topic alone for 5 minutes, using the same feedback for each
condition as Studies 1 and 2. They then completed the dependent measures individually
and in silence.
Dependent measures
The main dependent variable was a behavioural measure. Participants were asked to
vote in a fictitious ‘UK Youth Parliament Election’. They placed a vote for a candidate
of one of the following political parties: Labour, Conservative, the Liberal Democrats,
or the UK Independence Party (UKIP). Fictional male candidates were presented as
candidates for each party. A paragraph described each party’s immigration policy. The
policies described were genuine party statements taken from the respective parties’
programmes, and represented a continuum of political ideologies, from left- to rightwing. The Liberal Democrats were relatively pro-immigration. Labour and Conservative
policies were roughly equivalent centrist ones, stating the need for balancing the benefit
of immigration with control of the country’s borders. The UKIP policy was explicitly antiimmigration. We emphasize that these are actual policies of the main political parties
in Britain – the only aspects of the feedback that were fictional were the names of
the candidates and the election in which they were featured. Votes were subsequently
classified as hostile towards the out-group (UKIP vote: coded ‘1’) or not hostile towards
the out-group (not a UKIP vote: coded ‘0’).
The social validation scale used in Study 2 was expanded to six-items (␣ = .88), ‘I
feel that my opinions on immigrants are valid’; ‘I feel that my views on immigrants are
well-founded’; ‘My beliefs about immigrants are justified’; ‘I am certain that my views
on immigration are right’; ‘I feel that my opinions about immigration are shared by
many’; and ‘I feel that I have a justified opinion on immigration’. The same measure of
identification was included as in Study 1 (seven items, ␣ = .77), alongside the 10 item
generalized prejudice scale (␣ = .92).
Results and discussion
A Bernoulli sampling HLM analysis using a logit link function was conducted to assess
the magnitude of variation in UKIP votes (a binary measure) between conditions. This
analysis provides the probability that participants cast an anti-immigrant vote (i.e.,
the odds of voting UKIP rather than another party). Overall, the analysis showed
marginally significant between-condition differences, ␹ 2 (25) = 36.84, p = .06. Contrary
to Hypothesis (a), the main effect for type of reflection was not significant, ␥ = 0.11, logodds = 1.12, p = .85. There was however a main effect for topic of reflection, with more
UKIP votes overall when participants reflected upon the stereotype than when actions
The power of talk
207
were also considered, ␥ = 1.18, log-odds = 3.27, p = .05. The two-way interaction was
not significant ␥ = 0.94, log-odds = 2.55, p = .12.
Contrasts then tested the second hypothesis and other between-condition effects.
The topic of reflection main effect appeared to be driven by a highly significant increase
in probability of voting UKIP after the out-group stereotype discussion, which had the
most UKIP votes overall (54.5%) compared to the action discussion, which had the least
(10.5%; Table 4), ␥ = 2.12, log-odds = 8.33, p = .01, confirming Hypothesis (b) and
replicating the results of Study 2. There was no difference between the two individual
reflection conditions ␥ = 0.25, log-odds = 1.28, p = .78. There were no simple main
effects for type of reflection in the stereotype only ␥ = 1.05, log-odds = 2.84, p =
.18, or stereotype plus stereotype plus action conditions ␥ = −0.83, log-odds = .44,
p = .35.
Table 4. Frequency of UKIP votes, mean individual-level scores, and scale intercorrelations, in the
stereotype discussion (N = 21), action discussion (N = 18), stereotype reflection (N = 21), and action
reflection (N = 18) conditions, Study 3
Identification, generalized prejudice, and social validation scores were then analysed using HLM. Neither main effects nor two-way interactions were significant for
identification, social validation, or generalized prejudice (all p’s > .20), although
prejudice and validation were both significantly positively correlated with voting UKIP
(Table 4). Like in Study 2, an examination of the intra-class correlations between groups
in each condition indicates that there was least variability in UKIP votes between
groups in the stereotype only discussion condition ICC1 = 15.97. The stereotype plus
action discussion condition ICC1 = 57.87 and the stereotype only individual condition
ICC1 = 50.12 were similar. The highest variability was found in the stereotype plus
action individual condition ICC1 = 99.01.
Therefore, the results supported the second hypothesis, but not the first hypothesis.
In contrast to Hypothesis (a), although there were somewhat more UKIP votes when
participants engaged in a group discussion than when they reflected individually, this
was not a significant difference. The results did support the second hypothesis, however.
As in Study 2, discussion of the out-group stereotype over discussing of specific courses
of action produced an increase in anti-immigrant voting behaviour. There was least
208
Laura G. E. Smith and Tom Postmes
variability across groups within the stereotype discussion condition on the UKIP votes,
and far more variability when groups were asked to discuss action. However, in none
of the studies conducted so far could we control the content of the discussions. The
content is a key factor to control because groups could polarize merely on the basis of
the content of the information conveyed, irrespective of the group dynamic involved.
Study 4 was designed to investigate this possibility.
STUDY 4
This study examined whether or not the process of reaching a validating consensus on
the out-group stereotype is critical for formation of local discriminatory group norms.
The objective was to keep the information conveyed constant, whilst manipulating
whether or not a validating consensus on the out-group stereotype was achieved.
Therefore, participants watched a film of a scripted group discussion. There were
two versions of this discussion, both with the same arguments. However, the order
in which arguments were presented was varied, so that in one condition the discussion
appeared to converge, whilst in the other the discussion diverged. To our knowledge,
this is the first time that the process of reaching consensus has been manipulated. This
design enabled us to test whether the effect of social interaction on action intention
and identity content is driven by exposure to persuasive arguments alone, or by the
perception of the process of reaching a validating group consensus on the out-group
stereotype.
Hypotheses
We hypothesized that when the discussion ended in consensus (compared to the
discussion which ended in dissent), participants would (a) have greater intention to
participate in collective action and (b) this would be mediated by the extent to which
they perceived the discussion as normative.
Method
Participants and design
Participants (N = 63) were recruited via university e-mail lists and a social networking
website. Mean age was 26.79 years, and 47 participants were female. 90.5% stated their
nationality as British.4 This study used a one-factor (direction of consensus: consensus to
dissent vs. dissent to consensus vs. collective action base-rate) between-subjects design.
Participants were unpaid and naive as to the purposes of the study, and were randomly
assigned to either the consensus to dissent (N = 21), dissent to consensus (N = 20), or
control (N = 22) condition.
4 All
participants were included in the analyses regardless of their nationality. As removing non-British participants did not
make a qualitative difference to the results, their data were retained in order to avoid the reduction in statistical power that
comes from decreased sample size (22 participants for each cell is recommended to achieve sufficient power (␤ ⬎ 0.8) to
detect large effects; Faul, Erdfelder, Lang, & Buchner, 2007).
The power of talk
209
Stimuli and procedure
The study was conducted on-line. After informed consent was obtained, participants
were asked to state whether or not they were British, in order to help make national
identity salient. In the control condition, participants were informed that they were
taking part in a study on opinions towards immigration and simply answered the
questionnaire (which included only the relevant scales). In the experimental conditions,
participants were asked first to watch a 5 minute film of a group of British people (i.e., ingroup members) ostensibly discussing their opinions on immigration, and then answered
the questionnaire.
The film consisted of six British actors (three male, three female) who held a scripted
discussion about immigration in the UK. The statements in this script were based
upon the qualitative data reported in Study 1, therefore each explicit statement about
immigration was volunteered by previous participants. This meant that participants
were not exposed to any arguments that they could not potentially encounter in their
everyday lives, and the experimenter’s only involvement was to select a coherent and
representative set of statements about immigrants and order them so that they appeared
to progress logically. Although both films contained the same set of statements, the
order in which they were presented was systematically varied so as to show a discussion
either from consensus to dissent, or from dissent to consensus. There were two parts
to the discussion. In one part (X), the group was in dissent: both pro-immigration
arguments (e.g., economic stability, asylum) and anti-immigration arguments (e.g.,
welfare fraud, cultural clashes) were forwarded in equal measure. Vocal and body
language cues indicated disagreement within the group. In the other part (Y), there was
apparent consensus: only anti-immigration arguments were forwarded and vocal and
body language cues indicated majority consensus and social validation. In the dissent
to consensus condition, the film was edited to show (X) then (Y). In the consensus to
dissent condition, the film showed (Y) then (X). This gave the impression that the group
either did or did not reach consensus, respectively.5 Participants then completed the
questionnaire which consisted of standardized scales.
Dependent measures
Intergroup action intentions were measured with eight items (␣ = .86), ‘I would
participate in raising our collective voice to address the problems that immigration
brings’; ‘Immigration policy should be allowed to continue as it is now’ (reverse
scored); ‘I would do something to address the problems that immigration brings’;
‘I would encourage others to do something about immigration’; ‘I agree to have my
name added to a petition to raise awareness of the problems that immigration brings’;
‘I would participate in a demonstration against immigration’; and ‘I want to help to
decrease immigration’. A four-item scale measured to what extent participants were
able to perceive in-group norms form through the discussion (␣ = .83). We argued
that participants would be more able to infer norms from the discussion if it ended in
consensus relative to dissent. Items were, ‘The people on the film seem to hold normal
views’; ‘The views expressed in the film seemed excessive’ (reverse coded); ‘The people
5 This study was designed to be congruent with the previous studies which investigated the formation of norms for discriminatory
intergroup action. Therefore, we did not also manipulate reaching consensus on pro-immigration norms. Moreover, if the design
had included a positive consensus condition, we would no longer have been able to include exactly the same content in each
condition, introducing a confound.
210
Laura G. E. Smith and Tom Postmes
on the film hold the views of someone like me’; and ‘The views expressed in the film
seemed extreme’ (reverse coded). Finally, a six-item scale was included to measure
participants’ identification with the discussants on the film (␣ = .90; Ellemers et al.,
1999; see Study 1).
Results
A one-way analysis of variance (direction of consensus: dissent to consensus vs. collective
action base-rate vs. consensus to dissent) was conducted on action intention scores. The
condition effect was significant, F(2, 60) = 3.62, p = .03, ␩2p = .11. Contrasts showed
that there was significantly more action intention after watching the discussion which
ended in consensus (M = 4.25, SD = 1.45) than the discussion which ended in dissent
(M = 3.21, SD = 1.07) F(1, 60) = 7.24, p < .001, ␩2p = .11, as predicted. The base-rate
level of action intentions (M = 3.73, SD = 1.19) fell in the middle of the means in the
two experimental conditions, and it was not significantly different to either (both p’s >
.16).
A one-way ANOVA comparing two conditions (direction of consensus: consensus
to dissent vs. dissent to consensus) was conducted on scores from each of the
additional standardized scales. There was a significant condition effect on the norm scale,
F(1, 39) = 6.00, p = .02, ␩2p = .13. Participants considered the views expressed in the
discussion to be more normative when the group reached consensus (M = 4.74, SD =
1.22) than when the discussion ended in dissent (M = 3.79, SD = 1.26). The condition
effect on social validation was not significant, however F(1, 60) = 0.43, p = .52, ␩2p < .001
(Table 5).
There was a main effect of experimental condition on identification with the group
members in the discussion, F(1, 39) = 6.04, p = .02, ␩2p = .13, with significantly more
social identification after watching the discussion to consensus (M = 4.21, SD = 1.26)
than to dissent (M = 3.33, SD = 1.10). There was also a trend towards more endorsement
of the views expressed in the film when the discussion ended in consensus (M = 3.64,
SD = 1.26), compared to when it ended in dissent (M = 2.94, SD = 1.15), although
the main effect was marginal F(1, 39) = 3.44, p = .07, ␩2p = .08. These results together,
suggested that the process of reaching consensus informed identification and identity
content.
Table 5. Mean scores and inter-scale correlations in the consensus to dissent (N = 21) and dissent to
consensus (N = 20) conditions, Study 4
The power of talk
211
Norm
perception
.31†
.95*
Consensus
vs. Dissent
(1.04**)
.75, ns
Intergroup
action
intentions
Figure 3. Coefficients for the mediation model, Study 4 († p = .06; ∗ p < .05; ∗∗ p < .01).
Mediation
There were positive correlations between the results of the action intention scale and
perceptions of the norm r = .40, p < .001. In order to examine whether perceptions of
the norm could account for the effect of condition on action intentions, we conducted
a mediation analysis using Preacher and Hayes’ (2004) macro (Figure 3). The norm was
a marginally significant covariate, F(1, 38) = 3.84, p = .06, ␩2p = .09, reducing the
difference between the experimental conditions on action intention from F(1, 39) =
6.96, p = .01, ␩2p = .09 to F(1, 38) = 3.34, p = .08, ␩2p = .08, bootstrap confidence
interval = .02 to .74, number of samples = 1,000.
Discussion
The results of Study 4 show that, as hypothesized, (a) participants exhibited more action
intention after listening to the discussion which ended in consensus than the discussion
which ended in dissent; and (b) this was mediated by the extent to which participants
perceived the discussion as normative. Therefore, the results of Study 4 further suggest
that it was not the content of the group discussion per se that was driving the effect of
the manipulation, but the perception that the group reached a validating consensus on
the out-group stereotype. Furthermore, participants identified more with the discussants
when their discussion reached consensus than when it ended in dissent. Therefore, it
appears that a sense of shared identity with the discussants emerged as well as group
norms, that was premised upon the shared opinions that the discussants communicated.
GENERAL DISCUSSION
We conducted four studies to investigate the impact of discussing the societal stereotype
of a minority out-group on intergroup behavioural intentions. Overall, (a) group
discussion was more likely than individual rumination to focus on certain negative
aspects of the out-group stereotype, (b) participants were more likely to support
discriminating against the out-group when they discussed the out-group stereotype than
in other conditions, and (c) interaction had these profound effects because it validated
negative stereotypical perceptions of the out-group, contributing to the formation of a
norm which supported discriminatory behaviour.
212
Laura G. E. Smith and Tom Postmes
The power of talk: The implications
Although the studies were clearly framed within the intergroup context in Britain at the
time of the research, we argue that additional processes, over and above the effect of
the intergroup context, were occurring in these studies. The intergroup context and
topic (and therefore, arguably identity) salience was constant across the discussion and
individual reflection conditions. Therefore, we argue that the research presented here
demonstrates that group discussion uniquely impacted upon norm formation.
This contributes to the social identity literature by highlighting the important role
of group communication (contextualized within the intergroup context) in the development of norms for intergroup behaviour. Therefore, when investigating intergroup
behaviour, not only must the intergroup context be taken into account, but also how and
why social reality may be channelled through in-group dialogue (Reicher et al., 1997; see
also Postmes, Haslam, et al., 2005; Smith & Postmes, 2009). The present data also unify
past research which argued that (1) group discussion can increase group consensus on
stereotypes (e.g., Haslam, 1997); (2) group members’ agreement can increase individuals’
confidence in their beliefs (Festinger, 1950, 1954; McGarty et al., 1993); and (3) shared
stereotypes can provide a basis for intergroup behaviour (e.g., Stott & Drury, 2004).
Our focus on the role of intragroup processes in the development of norms does not
underplay the importance of historical and ideological factors (e.g., Subašić & Reynolds,
2009), the intergroup comparative context (e.g., Turner et al., 1987), or individual
cognitive processes (e.g., Greenwald & Banaji, 1995). Indeed, all of these aspects
contribute uniquely to an individuals’ subjective psychological ‘life space’ (Lewin, 1936).
The purpose of this research was to explore in particular the influence of intragroup
dynamics, not least because this issue tends to be underrepresented within this literature
(Paluck & Green, 2009, p. 354).
By exploring the issue of intragroup dynamics, we have shown that even a relatively
brief discussion can have a significant impact on participants’ responses. However,
future research should provide evidence that the effects could be a long-term normative
shift, rather than a temporary, contextual effect. After all, people may tune their
communications to their audience (e.g., Clark & Murphy, 1982) and the content of
that tuned communication can subsequently bias their memories of the message topic
(Echterhoff, Higgins, & Groll, 2005; for a review see Higgins, 1999). The mediating
effect of social validation (Study 2), increased shared cognition (Study 1), and the effect
of reaching consensus (Study 4) suggests that a normative shift was indeed occurring
in this research, rather than simply audience tuning. However, longitudinal evidence
would add weight to this claim.
One of the interesting features of the present research was that validating individuals’
views of the out-group was more easily achieved than reaching agreement on intergroup
actions. It may be that agreeing upon appropriate actions was more difficult than agreeing
on an out-group stereotype. Articulating support for what may be considered intergroup
discrimination may be constrained by self-presentational or additional identity concerns.
For example, discussion of action in relation to migrants may have invoked other
identities, such as political identities, which then highlighted the political problems
with endorsing action towards members of lower status groups. Increasing the salience
of political identities may have produced and emphasized disagreement within the
discussion group on appropriate actions. Consequently, this may have led to less
consensus and thereby undermined the general intergroup bias that resulted from
discussing only the stereotype. Discussing only the stereotypes may have been more
‘effective’ in promoting discriminatory norms because stereotypes (unlike specific
The power of talk
213
courses of action) are to some extent societally shared, and hence easier upon which to
reach consensus.
The study of the effect of in-group discussion on intergroup behaviour is relatively
embryonic. Perhaps inevitably at this stage, the research raises many questions. A further
challenge for future research is to determine whether the group dynamics documented
here can also contribute to improvements in intergroup relations. In theory, there is
no reason why the same principles operating to increase intergroup hostility could not
also be mobilized to reduce it. Thus, discussions among members of deprived groups
could equip them with the social solidarity and cognitive equipment to better resist
discrimination (see also Stott & Drury, 2004). Furthermore, discussion within higher
status groups could be equally productive, provided that a significant minority within that
group has to courage to voice dissent and undermine perceptions of social consensus.
Conclusion
The present research examines the consequences of group discussion for intergroup
discrimination. Discussion may validate negative out-group stereotypes and this could
contribute to hostile intergroup behaviour. This implies that the in-group provides
a forum within which the perceived appropriateness of intergroup behaviour is
negotiated. However, although the findings demonstrate that such debates may have
problematic social consequences, they simultaneously show that debates may only have
such consequences to the extent that the discussion is allowed to focus on the narrow
range of xenophobic prejudices upon which a majority can agree. The moment that
more nuanced and complex issues of intergroup relations are discussed, intragroup
debate may cease to be such a problem. Indeed, future research should explore the
conditions under which such nuanced discussions can provide integrative solutions for
intergroup relations.
References
Baron, R. S., Hoppe, S. I., Kao, C. F., Brunsman, B., Linneweh, B., & Rogers, D. (1996). Social
corroboration and opinion extremity. Journal of Experimental Social Psychology, 32, 537–
560. doi:10.1006/jesp.1996.0024
Clark, H. H., & Murphy, G. L. (1982). Audience design in meaning and reference. In J. F. Leny &
W. Kintsch (Eds.), Language and comprehension (pp. 287–299). Amsterdam: North-Holland.
Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological
Measurement, 20, 37–46. doi:10.1177/001316446002000104
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation
analysis for the behavioral sciences (3rd ed.). Hillsdale, NJ: Erlbaum.
Daily Mail (2009). ‘We got it wrong’: Embattled minister Phil Woolas finally admits Labour’s
immigration policy has failed. Retrieved from www.dailymail.co.uk/news/article-1181003/
We-got-wrong-Woolas-finally-admits-Labours-failed-immigration-policy.html
Echterhoff, G., Higgins, T., & Groll, S. (2005). Audience-tuning effects on memory: The role of
shared reality. Journal of Personality and Social Psychology, 89(3), 257–276. doi:10.1037/
0022-3514.89.3.257
Ellemers, N., Kortekaas, P., & Ouwerkerk, J. (1999). Self-categorisation, commitment to the group
and group self-esteem as related but distinct aspects of social identity. European Journal
of Social Psychology, 29, 371–389. doi:10.1002/(SICI)1099-0992(199903/05)29:2/3371::
AID-EJSP9323.0.CO;2-U
214
Laura G. E. Smith and Tom Postmes
Faul, F., Erdfelder, E., Lang, A., & Buchner, A. (2007). G∗ Power 3: A flexible statistical power
analysis program for the social, behavioural, and biomedical sciences. Behavior Research
Methods, 39(2), 175–191.
Festinger, L. (1950). Informal social communication. Psychological Review, 57, 271–282. doi:10.
1037/h0056932
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140.
doi:10.1177/001872675400700202
Fraser, C., Gouge, C., & Billig, M. (1971). Risky shifts, cautious shifts and group polarization.
European Journal of Social Psychology, 1, 7–29. doi:10.1002/ejsp.2420010103
Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition: Attitudes, self-esteem, and
stereotypes. Psychological Review, 102(1), 4–27. doi:10.1037/0033-295X.102.1.4
Gregor, N. (2005). Nazism, war and genocide: Essays in honour of Jeremy Noakes. Exeter:
University of Exeter Press.
Guimond, S., & Dambrun, M. (2002). When prosperity breeds intergroup hostility: The effects of
relative deprivation and relative gratification on prejudice. Personality and Social Psychology
Bulletin, 28, 900–912.
Haslam, S. A. (1997). Stereotyping and social influence: Foundations of stereotype consensus. In
R. Spears, P. J. Oakes, N. Ellemers & S. A. Haslam (Eds.), The social psychology of stereotyping
and group life (pp. 119–143). Oxford: Blackwell.
Haslam, S. A., Oakes, P. J., Reynolds, K. J., & Turner, J. C. (1999). Social identity salience and
the emergence of stereotype consensus. Personality and Social Psychology Bulletin, 25(7),
809–818. doi:10.1177/0146167299025007004
Haslam, S. A., Turner, J. C., Oakes, P. J., McGarty, C., & Reynolds, K. J. (1998). The group as a basis
for emergent stereotype consensus. European Review of Social Psychology, 28, 755–776.
doi:10.1002/(SICI)1099-0992(199809/10)28:5<755::AID-EJSP891>3.0.CO;2-Z
Higgins, E. T. (1999). Saying is believing effects: When sharing reality about something biases
knowledge and evaluations. In L. L. Thompson, J. M. Levine, & D. M. Messick (Eds.), Shared
cognition in organizations: The management of knowledge (pp. 33–49). Mahwah, NJ:
Erlbaum.
Ipsos MORI (2008). Attitudes towards immigration. Retrieved from www.ipsos-mori.com/
content/attitudes-towards-immigration.ashx
Jost, J. T., & Banaji, M. R. (1994). The role of stereotyping in system justification and the production
of false consciousness. British Journal of Social Psychology, 33, 1–27.
Kashima, Y., Fiedler, K., & Freytag, P. (2008). Stereotype dynamics: Language-based approaches
to the formation, maintenance, and transformation of stereotypes. London: Erlbaum.
Krull, J. L., & MacKinnon, D. P. (1999). Multilevel mediation modeling in group-based intervention
studies. Evaluation Review, 23, 418–444. doi:10.1177/0193841X9902300404
Lewin, K. (1936). Principles of topological psychology. London: McGraw-Hill Book.
Lewin, K. (1953). Studies in group decision. In D. Cartwright & A. Zander (Eds.), Group dynamics:
Research and theory (pp. 287–301). New York: Row, Peterson.
Luus, E. C. A., & Wells, G. L. (1994). The malleability of eyewitness confidence: Co-witness and
perseverance effects. Journal of Applied Psychology, 79, 714–723. doi:10.1037/0021-9010.
79.5.714
McGarty, C., Turner, J. C., Oakes, P. J., & Haslam, S. A. (1993). The creation of uncertainty in
the influence process: The roles of stimulus information and disagreement with similar others.
European Journal of Social Psychology, 23, 17–38. doi:10.1002/ejsp.2420230103
Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis: An expanded sourcebook
(2nd ed.). London: Sage.
Moscovici, S., & Zavalloni, M. (1969). The group as a polarizer of attitudes. Journal of Personality
and Social Psychology, 12, 125–135. doi:10.1037/h0027568
Myers, D. G., & Bishop, G. D. (1970). Discussion effects on racial attitudes. Science, 169(3947),
778–779. doi:10.1126/science.169.3947.778
The power of talk
215
Paluck, E. L., & Green, E. L. (2009). Prejudice reduction: What works? A review and assessment
of research and practice. Annual Review of Psychology, 60, 339–367. doi:10.1146/annurev.
psych.60.110707.163607
Petty, R. E., Briñol, P., & Tormala, Z. L. (2002). Thought confidence as a determinant of persuasion:
The self-validation hypothesis. Journal of Personality and Social Psychology, 82, 722–741.
doi:10.1037/0022-3514.82.5.722
Postmes, T., Haslam, S. A., & Swaab, R. (2005). Social influence in small groups: An interactive
model of social identity formation. European Review of Social Psychology, 16, 1–42. doi:10.
1080/10463280440000062
Postmes, T., Spears, R., Lee, T., & Novak, R. (2005). Individuality and social influence in
groups: Inductive and deductive routes to group identity. Journal of Personality and Social
Psychology, 89, 747–763. doi:10.1037/0022-3514.89.5.747
Preacher, K. J., & Hayes, A. F. (2004). SPSS and SAS procedures for estimating indirect effects
in simple mediation models. Behavior Research Methods, Instruments, and Computers, 36,
717–731.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data
analysis methods (2nd ed.). Thousand Oaks, CA: Sage.
Reicher, S., Hopkins, N., & Condor, S. (1997). Stereotype constructions as a strategy of influence. In
R. Spears, P. J. Oakes, N. Ellemers, & S. A. Haslam (Eds.), The social psychology of stereotyping
and group life (pp. 94–117). Oxford: Blackwell.
Ruscher, J. B., Hammer, E. Y., & Hammer, E. D. (1996). Forming shared impressions through
conversation: An adaptation of the continuum model. Personality and Social Psychology
Bulletin, 22, 705–720. doi:10.1177/0146167296227005
Sherif, M. (1956). Experiments in group conflict. Scientific American, 195(5), 54–58. doi:10.1038/
scientificamerican1156-54
Smith, L. G. E., & Postmes, T. (2009). Obstructed ingroup advancement and the development of
norms which eliminate the positive–negative asymmetry effect. European Journal of Social
Psychology, 39, 130–144. doi:10.1002/ejsp.464
Stangor, C., Sechrist, G. B., & Jost, J. T. (2001). Changing racial beliefs by providing consensus information. Personality and Social Psychology Bulletin, 27, 486–496. doi:10.1177/
0146167201274009
Stoner, J. A. F. (1968). Risky and cautious shifts in group decisions: The influence of widely held
values. Journal of Experimental Social Psychology, 4, 442–459. doi:10.1016/0022-1031(68)
90069-3
Stott, C., & Drury, J. (2004). The importance of social structure and social interaction in stereotype
consensus and content: Is the whole greater than the sum of its parts? European Journal of
Social Psychology, 34, 11–24. doi:10.1002/ejsp.183
Subašić, E., & Reynolds, K. J. (2009). Beyond practical reconciliation: Intergroup inequality and
the meaning of non-indigenous identity. Political Psychology, 30(2), 243–267. doi:10.1111/j.
1467-9221.2008.00690.x
Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict. In W. G. Austin &
S. Worchel (Eds.), The social psychology of intergroup relations (pp. 33–47). Monterey, CA:
Brooks/Cole.
Thompson, M. S., Judd, C. M., & Park, B. (2000). The consequences of communicating social
stereotypes. Journal of Experimental Social Psychology, 36, 567–599. doi:10.1006/jesp.1999.
1419
Turner, J. C. (1991). Social influence. Milton Keynes: Open University Press.
Turner, J. C., Hogg, M. A., Oakes, P. J., & Reicher, S. D. (1987). Rediscovering the social group: A
self-categorisation theory. Oxford: Blackwell.
Received 3 February 2009; revised version received 11 March 2010